Re: 2.5.22: FB_VESA - early crash in fbcon_cursor()

[Paul Mundt <lethal@ChaoticDreams.ORG>]

[-- Attachment #1: Type: text/plain, Size: 1767 bytes --]

Hi Martin,

On Tue, Jun 18, 2002 at 12:31:29PM +0200, Martin Diehl wrote:
> no luck here getting 2.5.22 to boot on my ob800 with vesafb enabled.
> Same with 2.5.21. Last one working was 2.5.16 but I haven't tried 17-20.
> Box oopses due to NULL-pointer dereference during inital fbdev setup.
> 
> config, dmesg + decoded dump from serial console below.
> 
<snip>

> Code: 8b 40 18 85 c0 74 4a 66 8b 7f 2c 89 ea 66 89 bb e4 00 00 00 
> Error (Oops_bfd_perror): set_section_contents Bad value
> 
> >>EIP; c0192baf <fbcon_cursor+6f/200>   <=====
> Trace; c0177281 <hide_cursor+81/90>
> Trace; c017a79c <vt_console_print+8c/310>
<snip>

Looks like the dispsw isn't being set and you're running into the NULL
dereference in fbcon_cursor() upon trying to dereference it.. it looks like
fbgen.c is the culprit here, as it never sets display->dispsw if we aren't in
24-bpp or have FBCON_HAS_ACCEL set..

James, what's the point of th FBCON_HAS_ACCEL ifdef? It looks like all the
accel wrapper code does is provide a wrapper to the fillrect, imageblit, and
copyarea routines -- if the driver doesn't have accelerated ones to provide
for itself, it just uses the cfb_fillrect/imageblit/copyarea as a fallback,
thus it should _always_ be safe to call them.

If that's not the case, we'll have to re-introduce the FBON_CAS_CFBx
brain-damage in gen_set_disp() to keep dispsw happy.

Regards,

-- 
Paul Mundt <lethal@chaoticdreams.org>

--- linux-fbdev-2.5/drivers/video/fbgen.c	Tue Jun 18 11:37:46 2002
+++ linux-fbdev-2.5/drivers/video/fbgen.c	Tue Jun 18 11:37:59 2002
@@ -452,10 +452,9 @@
 #endif
 	}
 
-#ifdef FBCON_HAS_ACCEL
 	display->scrollmode = SCROLL_YNOMOVE;
 	display->dispsw = &fbcon_accel;
-#endif
+
 	return;
 }
 

[-- Attachment #2: Type: application/pgp-signature, Size: 240 bytes --]